Project specifications
Outline
Describe the underlying patient demographics within out mitigable-activity sub-cohorts by:
- Age and sex
- Ethnicity and deprivation
Present activity totals and time-series trends for the combined mitigable-activity cohort and mitigator sub-cohorts.
Measures include:
- Counts of patients, admissions and bed days:
- In combined mitigable-activity cohort
- By mitigator sub-cohorts
- By ICB
- Cohort activity as a share of all admissions
Explore variation in age- and sex-adjusted admission rates to:
- Examine the range of activity by ICB
- Assess whether systems in place are particularly good or bad at treating mitigable activity
Additional analysis: Incorporate underlying population need and/or disease prevalence (weighting by deprivation or regression controlling for population variables)
Apply survival analysis techniques to assess differences in care outcomes by patient groups and ICBs. Care outcomes include:
- Readmission post-acute inpatient care
- Motality post-acute inpatient care
Additional analysis:
- Summary of care services/types that patients from mitigable cohorts are in contact with in the year prior to death.
- Comparison of location of death in mitigable patients that recieved acute inpatient care in the year prior to death.
Display the overlap of patients in mitigable cohorts by combined and sub-cohorts.
Outputs include:
- Underlying SQL queries
- Code
- Data extract
- Quarto report output
Ranked table of activity in each mitigator with activity counts and proportion of activity
Cohort definition
In keeping with the wider context of this work, we have recreated mitigable activity cohorts used elsewhere in the New Hospital Programme (NHP) Demand and Capacity Model. Our extracts differ primarily in datasource, where the NHP model is built using Hosptial Episode Data (HES) while we have accessed Secondary Uses Service (SUS) data via the National Commissioning Data Repository (NCDR) portal. As such, we have included data between April 2018 and August 2024 and include activity from across England.
Full SQL queries used to define our patient cohorts can be found on the project github repository: sql_queries.
Details of the original NHP queries can be found here: NHP mitigators.
Admissions included in our emergency elderly cohort are all emergency admissions in patients aged 75 and older (Admission_Method LIKE “2%” - details here).
Admissions included in our Frail cohort are all emergency admissions in patients aged 65 and older that include a Spell_Primary_Diagnosis code indicating frailty as the cause of the admission. This is different to the NHP frailty definition which focuses only on patients aged 75 years and older.
Expand To view ICD10 frailty diagnosis codes:
LEFT([Spell_Primary_Diagnosis],3) IN ( ‘F00’,‘G81’,‘G30’,‘I69’,‘R29’,‘R39’,‘F05’,‘W19’, ‘S00’,‘R31’,‘B96’,‘R41’,‘R26’,‘I67’,‘R56’,‘R40’, ‘T83’,‘S06’,‘S42’,‘E87’,‘M25’,‘E86’,‘R54’,‘Z50’, ‘F03’,‘W18’,‘Z75’,‘F01’,‘S80’,‘L03’,‘H54’,‘E53’, ‘Z60’,‘G20’,‘R55’,‘S22’,‘K59’,‘N17’,‘L89’,‘Z22’,
‘B95’,‘L97’,‘R44’,‘K26’,‘I95’,‘N19’,‘A41’,‘Z87’,
‘J96’,‘X59’,‘M19’,‘G40’,‘M81’,‘S72’,‘S32’,‘E16’,
‘R94’,‘N18’,‘R33’,‘R69’,‘N28’,‘R32’,‘G31’,‘Y95’,
‘S09’,‘R45’,‘G45’,‘Z74’,‘M79’,‘W06’,‘S01’,‘A04’,
‘A09’,‘J18’,‘J69’,‘R47’,‘E55’,‘Z93’,‘R02’,‘R63’,
‘H91’,‘W10’,‘W01’,‘E05’,‘M41’,‘R13’,‘Z99’,‘U80’,
‘M80’,‘K92’,‘I63’,‘N20’,‘F10’,‘Y84’,‘R00’,‘J22’,
‘Z73’,‘R79’,‘Z91’,‘S51’,‘F32’,‘M48’,‘E83’,‘M15’,
‘D64’,‘L08’,‘R11’,‘K52’,‘R50’ )
Our falls cohort included any patient aged 65 years or older that had an emergency admission and a primary diagnosis indicating a fall (expliticly or implicitly).
Expand To view ICD10 falls diagnosis codes:
Explicit: Spell_Primary_Diagnosis LIKE ‘W[01][0-9]%’
Implicit - fractures: ( Spell_Primary_Diagnosis LIKE ‘M48[45]%’ OR Spell_Primary_Diagnosis LIKE ‘M80[0-589]%’ OR Spell_Primary_Diagnosis LIKE ‘S22[01]%’ OR Spell_Primary_Diagnosis LIKE ‘S32[0-47]%’ OR Spell_Primary_Diagnosis LIKE ‘S42[234]%’ OR Spell_Primary_Diagnosis LIKE ‘S52%’ OR Spell_Primary_Diagnosis LIKE ‘S620%’ OR Spell_Primary_Diagnosis LIKE ‘S72[0-48]%’ OR Spell_Primary_Diagnosis LIKE ‘T08X%’ ) AND Spell_Primary_Diagnosis NOT LIKE ‘[VWXY]%’
Implicit - Tendency to fall: Spell_Primary_Diagnosis LIKE ‘R296%’
End of life spells were identified by filtering admissions on Dischare_Method = 4 (Outcome - Patient died. Details here) where the length of stay (Der_Spell_LoS) was less than 14 days and no procedure was undertaken (Der_Procedure_All IS NULL).
Patient characteristics
Demographics
In both our emergency elderly and frail patient groups, it is apparent that within the Black and Asian ethnic groups, the majority of patients are clustered in the most deprived IMD deciles (particularly deciles 2-4). However, when considering emergency elderly admissions, in larger group of White British patients, the proportion of patients gradually increases with affluence.
Most frequent diagnoses
Most frequent procedures
Size and describe
The following section will describe mitigable-activity in our combined cohort and mitigator-specific sub-cohorts.
We have created our data set by applying existing strategies developed under the New Hospitals Program (NHP) that filter patient-level hospital activity data (Secondary Uses Service data - SUS) which center on patient groups and pathways which might reasonably be suitable for treatment in the community.
The following patient groups have been identified with such mitigation strategies:
- Admissions in frail elderly patients (65 years +)
- Emergency admissions in over 75’s
- End-of-life admissions: spells that ended with the patient dying and lasted less that 14 days
- Admissions post-falls (slips, trips and falls)
Combined mitigable activity cohort
Activity in the combined data set has gradually returned to pre-pandemic levels. While the increase in post-pandemic growth in spells has been gradual between 2021 and 2024, the length of time a person from our cohort has stayed in hospital has risen quickly to surpass pre-pandemic levels. The graph above suggests a change in the average length of stay particularly between 2021 and 2023, after which bed-days associated with our activity have plateaued.
Is the change in bed days linked to an easing of pandemic-related practices around length of treatment and/or discharge?
We contextualise the activity identified in our cohort alongside wider NHS-funded healthcare delivery. The combined sum of admissions from our emergency eldery, frail, falls and end-of-life cohorts accounts to approx 11% of total admissions. The trend varied most significantly during the COVID-19 pandemic and has since returned to pre-pandemic levels, allowing for consistent seasonal peaked during the winter months.
Sub-cohorts
By splitting our activity according to mitigation strategy, we can compare trends in patient group. The volume of activity by cohort differs significantly with Emergency Elderly and Frail activity accounting for the larger shares of total admissions and bed days.
Though the post-pandemic trend in frail activity is growing at a steady rate (from c.90,000 per month in 2022 up to c.100,000 per month since), there has been a recent stepped increase in emergency elderly admissions in the last year (stable around 140,000 per month between 2021-23 but increased to 160,000 since start of 2024). For both of these cohorts, the length of stay associated with these admissions has grown substantially and is higher than pre-pandemic levels in the emergency elderly cohort.
Similarly, the length of stay associated with falls admissions has surged while the underlying activity has remained stable.
Excluding pandemic related surges in end of life care in secondary settings, trends in activity and bed days are stable and correlated.
When we separate our patient cohorts, we’re reminded of the comparative differences in scale. The emergency elderly and frail cohorts account for approximately 10 and 6.5 per cent of total inpatient hospital activity respectively, while falls and end of life care are both below 1% total admissions. All cohorts saw increases in proportion during the pandemic as activity in other patient groups reduced.
ICB trends
In absolute volume terms, ICB’s differ significantly in the number of mitigable activity undertaken in secondary care settings though this may largely be a function of differences in underlying population distribution of elderly people.
The trends in mitigable activity by ICB vary significantly.
Data quality issues are apparent in a handful of ICB’s:
- Frimley,
- Shropshire,
- Telford and Wrekin,
- Cambs & Peterborough and
- Dorset.
While some ICB’s display sustained reductions in mitigable admissions:
- South West London,
- West Yorkshire and
- Lincolnshire.
Others show significant growth in activity:
- Devon,
- Staffordshire & Stoke-on-Trent and
- South Yorkshire for example
Are boundary changes impactful here?
Share of total activity in ICB
Below we display the range of variation in the mitigable activity as a proportion of total admissions by ICB. Each boxplot represents monthly proportions and particularly high or low outlier ICB’s are identified.
Use variation
To account for differences in the underlying number of elderly patients in each ICB, we standardised admission rates by age and sex to assess activity by ICB, if the age and sex structure of each ICB mirrored that of England (mid 2023).
Admission rates range from 25-55 admissions per 1,000 population between our ICB’s during our data collection period. While the general trend in admission rates is increasing, there is significant variation between ICB’s over the last 5 years, having accounted for age and sex structure.
Admission rate by sub-cohort
The overall trends in Emergency elderly and Frail patients are similar as the frail cohort is largely a sub-set of the emergency elderly group; increasing admission rates are seen in the post-pandemic period in both. While the trend in falls admission rates is more stable, more statistical outliers exist. The trend in end of life admission rates is reducing post-pandemic but has not returned to pre-pandemic levels and considerable variation by ICB is present where it wasn’t before the pandemic.
To assess the geographic distribution of admission rates we split our data into quintiles. Quintile 1 includes the ICB’s with the lowest 20% admission rates, while quintile 5 includes ICB’s with the highest 20%.
Visually we can identify that ICB’s in the higher admission rate quintiles for emergency elderly and frail cohorts are similar and cluster in the northern and midlands geographies. Where are ICB’s with the highest falls admission rates are located around London and ICB’s most often utilising secondary settings for end of life care are seen in the east, around Bristol, Worcester and Birmingham.
Alternatively we can visualise admission rates in our sub-cohorts by ICB using the below bar chart.
Or we can compare cohort activity by ICB using an average Length of Stay measure:
It is important to note however, there are limitations is using length of stay as a measure by which to compare ICB’s. Specifically :
- The range in length of stay values is small across most cohorts so the distinction between quintiles is often +/- a single day in average length of stay.
- The underlying data masks smaller differences in treatment length because stays are rounded to whole numbers.
- Similarly, presenting average length of stay does not illustrate the range of values within ICB’s that may be focused on specific populations or care types.
- Finally, presenting an annual average figure must be considered alongside data quality issues that affect only part of 2023 (the data collection period in question for the above graph) - Frimley ICB for example, is shown to have significant data quality issues affecting the later half of 2023, which need to be taken into account when interpreting the above visualisation.
Funnel plots
If we revert to considering admission rates per 1,000 population by patient sub-cohort and ICB, below we have identified the ICB’s that sit outside (+/-) 1 standard deviation from the mean admission rate and have plotted them according to underlying population size.
For frail, emergency elderly and falls cohorts, we note that those ICB’s that are demonstrating particularly low admission rates, they are often clustered in the lower left quadrant of the graph, indicating smaller populations (around 1,000,000 residents). However, the inverse is true for patients who died in hospital within 14 days of admission, where by the ICB’s with higher admission rates are more likely to have smaller resident populations.
2x2 plots
Finally when considering variation in use of acute care, we assess ICB admission rates in our patient cohorts against other cohorts to attempt to demonstrate the trade-offs at play for the system. Click through tabs for plot description:
- Frail x Emergency elderly: The cohorts are highly correlated (as the frail cohort is largely a subset of the emergency elderly cohort). ICB’s tend to be positioned similarly on both measures - i.e. if an ICB has a high admission rate for emergency elderly, they tend to have high admission rates in the frail cohort also.
- Frail x Falls: There is less evidence of correlation but almost all ICB’s that have admission rates below 1 standard deviation in the falls cohorts also are below 1 standard deviation for frail admissions (lower left quadrant).
- Frail x End of life: There is no evidence of a link between an ICB’s admission rate in end of life admissions and frail patients.
- Emergency elderly x Falls: Correlation is demonstrated again when considering falls and emergency elderly however the 4 ICB’s with the highest admission rates for falls are within 1 standard deviation of the mean emergency elderly admission rate and the 5 of the highest 7th emergency elderly rates are similarly closer to average for the falls cohorts.
- Emergency elderly x End of life: No evidence of correlation between end of life and emergency elderly admission rates with a number of ICB’s position far from average on 1 but near the average on the other.
- Falls x End of life: Similarly, no evidence of correlation between end of life and falls admission rates with a number of ICB’s position far from average on 1 but near the average on the other.
Variation in outcomes
Furthering our assessment of variation in treatment of these mitigable patient populations, we analyse variation in selected patient outcomes; readmission and mortality.
This sub-strand of our descriptive analysis aims to investigate the risk of death or subsequent readmission in patients who have had an emergency hospital admission. We apply survival analysis techniques to help us understand what factors influence the risk of death or readmission, including determining variability between ICB area.
Full survival analysis output found here: https://the-strategy-unit.github.io/Community_Strategies/community_strategy_survival_analysis.html
Mitigator overlap
Finally, explore the correlation and overlap between and within the cohorts included within the Community Strategies analysis.
Full analysis on overlap found here: https://the-strategy-unit.github.io/Community_Strategies/identifying_overlap_between_cohorts.html
Contact
If you have any questions or comments regarding any of the above analysis please email strategy.unit@nhs.net citing the NHSE Community Services Analysis.